Soil Moisture Sensing Using Zoomed-in Camera Images at Close Proximity

Soil moisture is a critical variable that influences various terrestrial processes and has significant implications for environmental science, land management, disaster mitigation, and civil engineering. However, current methods for soil moisture sensing often involve costly data collection procedures or high-end equipment, presenting a challenge for resource-constrained settings. This research proposes a cost-effective, accessible soil moisture sensing approach using machine learning algorithms to analyze soil images captured by a common mobile camera coupled with a high-magnification zoom lens and a FoldScope. This paper details the theory and implementation of a Support Vector Machine model for soil moisture prediction, utilizing the Scikit-Learn library for Python. We evaluated our model across three distinct soil types-loam soil, sand, and garden soil-under varying controlled moisture conditions. The results demonstrate that while the high-magnification zoom lens method achieved high accuracy across all soil types, particularly for sand, the FoldScope method, although less precise, could provide a viable, low-cost alternative for soil moisture sensing. These findings highlight the potential of machine learning and accessible imaging technologies in developing efficient, cost-effective soil moisture sensing techniques.